ai-portland-rocketleague

Materials/code for a short (<10m) presentation for an AI Portland event.

Here are the Presentation Slides

Overview

Rocket League is a fast-paced game that demands quick reflexes, precise control, and strategic team play. This project explores two primary approaches to integrating game mechanics with AI:

• Mechanics Feedback: Anthropic's Claude Sonnet 3.5 model was used to provide feedback on player mechanics during freeplay.
• Replay Prompt: OpenAI's Assistant API was used for a replay prompt during replays.

1. DribbleCoach : Mechanics Feedback

• Description: Textual feedback of ground and air dribble mechanics using Anthropic's Claude during freeplay.
  • Identifies and tracks the mechanical skill of ground and air dribbling.
  • Offers simple suggestions on optimal timing, positioning, and ball control.
• ToDo:
  • Fine-tune air dribbling tracking.
  • Provide flick feedback.

2. ReplayAssistant : Replay Prompt

• Description: Extracts replay data and creates an OpenAI assistant console available on the current replay.
• ToDo:
  • Improve the usability of the prompt
  • linking to prompt bidirectional communicaiton on current frame/timestamp/location
  • annotations/screenshot sharing like the replay plugin)

Objectives

• Improve player mechanics and strategic understanding.
• Foster better teamwork and communication within teams.
• Deliver actionable, easy-to-understand insights to players of all skill levels.
• Advance the use of AI in gaming to create a more immersive and educational experience.

Similar and Related Services/Projects

• polr
• trophi.ai
• bakkesmod
• carl
• rlgym
• replay review plugin
• ballchasing.com
• rocketleague.tracker.network/

License

This project is licensed under the Apache 2 License.